Fred,
There are alternatives....
Probably, the best alternative is for the tunnel ingress router to tunnel
ingress router to discover the PMTU to the egress. When the tunnel ingress
router receives a packet that is so large that it cannot be forwarded through
the tunnel, it discards the packet and sends an ICMP PTB to the packet's
originator. The packet's originator then modifies its sending behavior based
upon its new estimate of the PMTU associated with the destination.
So, for the purposes of MTU management, the tunnel is just another link.
Ron
> -----Original Message-----
> From: Templin, Fred L [mailto:[email protected]]
> Sent: Wednesday, July 10, 2013 10:59 AM
> To: Ronald Bonica; Doug Barton
> Cc: [email protected]
> Subject: RE: Meta-issues: On the deprecation of the fragmentation
> function
>
> Hi Ron,
>
> > -----Original Message-----
> > From: [email protected] [mailto:[email protected]] On Behalf
> > Of Ronald Bonica
> > Sent: Tuesday, July 09, 2013 5:33 PM
> > To: Doug Barton
> > Cc: [email protected]
> > Subject: RE: Meta-issues: On the deprecation of the fragmentation
> > function
> >
> > Doug,
> >
> > Let's see if we can find some common ground.
> >
> > Assume that the IETF is considering a new protocol that doesn't run
> > over TCP. In order to deal with MTU issues, the new protocol must do
> > one of the following:
> >
> > a) implement PLMTUD or PMTUD
> > b) restrict itself to sending PDUs so small that when they are
> > encapsulated in an IPv6 header, the resulting packet will not exceed
> > 1280 bytes
> > c) rely on IPv6 fragmentation
> >
> > Is there ever a reason why c) is better than a) or b). For that
> > matter, is c) ever an acceptable solution?
>
> Fragmentation at a tunnel ingress router is unavoidable. Proof:
>
> - a tunnel configures a 1280 MTU
> - When its packets are encapsulated they emerge as (1280 + HLEN)
> (the length of the encapsulating headers)
> - the tunnel crosses a 1280 link somewhere in the path to the egress
> - the packet is dropped with a PTB signal sent back
> - the ingress now has two choices: 1) start fragmenting, 2) quit.
>
> Thanks - Fred
> [email protected]
>
> > Ron
> >
> > > -----Original Message-----
> > > From: Doug Barton [mailto:[email protected]]
> > > Sent: Tuesday, July 09, 2013 2:41 PM
> > > To: Ronald Bonica
> > > Cc: [email protected]
> > > Subject: Re: Meta-issues: On the deprecation of the fragmentation
> > > function
> > >
> > > On 07/09/2013 11:12 AM, Ronald Bonica wrote:
> > > > Doug,
> > > >
> > > > It might be interesting to revisit what we mean by deprecating
> > > > IPv6
> > > fragmentation....
> > > >
> > > > It means that the IETF will not approve any new protocols that
> > > > rely
> > > upon IPv6 fragmentation. Nothing more, nothing less.
> > >
> > > Thanks for clarifying. FWIW, I understand what is being proposed,
> > > and
> > I
> > > still think it's a bad idea.
> > >
> > > Doug
> > >
> > >
> >
> >
> >
> > --------------------------------------------------------------------
> > IETF IPv6 working group mailing list
> > [email protected]
> > Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> > --------------------------------------------------------------------
>
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
